Simple comparative benchmarks

[ivirshup]

We should have a simple comparative benchmark (with prose) comparing using AnnData and our on disk format to alternative structures.

Initial measurements could look at:

• IO time
• IO memory
• On disk size
• In memory size

Alternatives to compare against include:

• Serialized R object (SeuratObject, maybe also SingleCellExperiment)
• Loom

Variables to consider here include file format (hdf5 vs zarr), chunking, and compression.

[ivirshup]

Good resources to use here include:

• Benchmarking chapter from Jake VanderPlas' Python Data Science HandBook
• loompy docs
• Seurat docs